feat: docsgpt backup and docsgpt restore

`docsgpt backup` writes one archive holding a pg_dump of the database, a tar
of each data volume and a manifest of what it came from; `docsgpt restore`
puts it back over an install. The settings file is left out unless
--with-settings asks for it, since it holds the install's secrets, and a
backup taken with a newer DocsGPT is refused without --force.

The volume tars go through the image the install already runs, so a backup
pulls nothing extra, and compose calls can now redirect stdout and stdin so
the dump never passes through this process.

docs/content/Deploying/Docker-Deploying.mdx

@@ -79,6 +79,36 @@ the questions again.
 | `docsgpt down` | Stop the stack; data and settings stay |
 | `docsgpt uninstall [--purge]` | Remove the containers; `--purge` also deletes the settings and all data |
 
+### Backups
+
+`docsgpt backup` writes one archive holding a dump of the database and a tar of
+each data volume (`indexes`, `inputs`, `vectors`):
+
+```bash
+docsgpt backup                      # into <stack>/backups
+docsgpt backup --out /mnt/backups   # somewhere else, e.g. a mounted disk
+```
+
+The archive does **not** include `.env`, because that file holds the install's
+secrets. `docsgpt backup --with-settings` puts it in, for when the archive
+itself is stored somewhere private. Keep `.env` safe separately otherwise: the
+database password in it is what an existing Postgres volume expects.
+
+Restoring replaces the data in an install:
+
+```bash
+docsgpt restore ~/.docsgpt/server/backups/docsgpt-20260916-120000.tar.gz
+```
+
+It asks first, then stops the stack, puts the volumes and the database back, and
+starts DocsGPT again. `--yes` skips the question for scripts. A backup taken
+with a newer DocsGPT is refused, since its data may not fit this version's
+schema; upgrade first, or pass `--force` if you know the two match.
+
+The Postgres data directory itself is not archived: the dump is the database
+backup, and copying a directory Postgres is writing to would capture a torn
+copy. Caddy's certificates are not archived either, as it obtains them again.
+
 More `docsgpt up` options: `--port`, `--docling` (the image with the docling
 parser engine and OCR), `--image-tag develop` (follow the `main` branch) and
 `--adopt` (manage a stack you started from the standalone Compose file in

docs/content/changelog.mdx

@@ -11,6 +11,16 @@ The notable changes in each release. Every release on GitHub also carries
 [auto-generated notes](https://github.com/arc53/DocsGPT/releases) listing every merged pull
 request, and [Upgrading](/upgrading) covers the steps an existing deployment has to take.
 
+## Unreleased
+
+### Back up and restore an install
+
+`docsgpt backup` writes a dump of the database and a tar of each data volume into one archive, and
+`docsgpt restore <archive>` puts them back. The settings file is left out unless
+`--with-settings` asks for it, since it holds the install's secrets, and a backup from a newer
+DocsGPT is refused unless you pass `--force`. See
+[Backups](/Deploying/Docker-Deploying#backups).
+
 ## 0.21.0
 
 ### Install with one command

docsgpt/cli.py

@@ -230,6 +230,18 @@ def _add_deploy_commands(commands) -> None:
     uninstall.add_argument("-y", "--yes", action="store_true", help="do not ask for confirmation")
     uninstall.add_argument("--purge", action="store_true", help="also delete the settings and all data")
 
+    backup = stack_command("backup", "backup", "write a backup of the database and the uploaded data")
+    backup.add_argument("--out", help="directory for the archive (default: <stack>/backups)")
+    backup.add_argument("--with-settings", action="store_true",
+                        help="include .env in the archive; it holds this install's secrets")
+
+    restore = stack_command("restore", "restore", "restore a backup over this install")
+    restore.add_argument("archive", help="the .tar.gz written by `docsgpt backup`")
+    restore.add_argument("-y", "--yes", action="store_true", help="do not ask for confirmation")
+    restore.add_argument("--force", action="store_true", help="restore a backup taken with a newer DocsGPT")
+    restore.add_argument("--timeout", type=int, default=300,
+                         help="seconds to wait for the API afterwards (default: 300)")
+
     env = stack_command("env", "env", "show, get or set the stack's settings")
     env_actions = env.add_subparsers(dest="env_action", metavar="<action>")
     get = env_actions.add_parser("get", help="print one setting")

docsgpt/deploy/backup.py

@@ -0,0 +1,107 @@
+"""The archive ``docsgpt backup`` writes and ``docsgpt restore`` reads.
+
+One gzipped tar holds a SQL dump of the database, a tar per data volume, and a
+manifest saying which version and image the backup came from. The settings file
+is left out unless it is asked for: it holds the secrets.
+
+``postgres_data`` is not tarred, because the dump is the database backup and a
+copy of a running data directory would be a torn one. Caddy's volumes are left
+out too: they hold certificates it obtains again on the next start.
+"""
+
+from __future__ import annotations
+
+import io
+import json
+import tarfile
+import time
+from collections.abc import Mapping
+from datetime import datetime
+from pathlib import Path
+from typing import Optional
+
+from docsgpt.deploy.docker import DeployError
+
+FORMAT = 1
+MANIFEST = "manifest.json"
+DUMP = "database.sql"
+SETTINGS = "settings.env"
+VOLUME_DIR = "volumes"
+DATA_VOLUMES = ("indexes", "inputs", "vectors")
+
+
+def archive_name(when: datetime) -> str:
+    """The file name for a backup taken at ``when``."""
+    return f"docsgpt-{when.strftime('%Y%m%d-%H%M%S')}.tar.gz"
+
+
+def volume_member(name: str) -> str:
+    """Where a volume's tar sits inside the archive."""
+    return f"{VOLUME_DIR}/{name}.tar"
+
+
+def write_archive(
+    path: Path,
+    *,
+    dump: Path,
+    volume_tars: Mapping[str, Path],
+    manifest: Mapping[str, object],
+    settings: Optional[Path] = None,
+) -> None:
+    """Write the backup archive; the manifest is added last so a truncated file has none."""
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with tarfile.open(path, "w:gz") as archive:
+        archive.add(dump, arcname=DUMP)
+        for name, tar_path in sorted(volume_tars.items()):
+            archive.add(tar_path, arcname=volume_member(name))
+        if settings is not None:
+            archive.add(settings, arcname=SETTINGS)
+        body = json.dumps(dict(manifest), indent=2).encode("utf-8")
+        info = tarfile.TarInfo(MANIFEST)
+        info.size = len(body)
+        info.mtime = int(time.time())
+        info.mode = 0o600
+        archive.addfile(info, io.BytesIO(body))
+
+
+def read_manifest(path: Path) -> dict:
+    """The archive's manifest, or a DeployError naming what is wrong with the file."""
+    if not path.is_file():
+        raise DeployError(f"{path} does not exist")
+    try:
+        with tarfile.open(path, "r:gz") as archive:
+            member = archive.extractfile(MANIFEST)
+            if member is None:
+                raise KeyError(MANIFEST)
+            manifest = json.loads(member.read().decode("utf-8"))
+    except (tarfile.TarError, KeyError, ValueError, OSError) as exc:
+        raise DeployError(f"{path} is not a DocsGPT backup: {exc}") from exc
+    if not isinstance(manifest, dict) or "volumes" not in manifest:
+        raise DeployError(f"{path} is not a DocsGPT backup: its manifest is missing what to restore")
+    return manifest
+
+
+def extract(path: Path, destination: Path) -> None:
+    """Unpack the archive into ``destination`` (data filter: no paths outside it, no devices)."""
+    with tarfile.open(path, "r:gz") as archive:
+        archive.extractall(destination, filter="data")
+
+
+def _parts(version: str) -> tuple[int, ...]:
+    numbers = []
+    for chunk in str(version).split("."):
+        digits = "".join(character for character in chunk if character.isdigit())
+        numbers.append(int(digits) if digits else 0)
+    return tuple(numbers)
+
+
+def check_version(manifest: Mapping[str, object], current: str, force: bool) -> None:
+    """Refuse a backup from a newer DocsGPT: its data may not fit this version's schema."""
+    taken_with = str(manifest.get("version") or "")
+    if force or not taken_with:
+        return
+    if _parts(taken_with) > _parts(current):
+        raise DeployError(
+            f"this backup is from DocsGPT {taken_with}, newer than the installed {current}. "
+            "Upgrade first with `docsgpt upgrade`, or pass --force to restore it anyway."
+        )

docsgpt/deploy/commands.py

@@ -9,6 +9,7 @@ import re
 import shutil
 import subprocess
 import sys
+import tempfile
 import webbrowser
 from collections.abc import Callable, Mapping
 from dataclasses import dataclass
@@ -16,6 +17,7 @@ from datetime import datetime, timezone
 from pathlib import Path
 from typing import Any, Optional
 
+from docsgpt.deploy import backup as backup_format
 from docsgpt.deploy import envfile, stack
 from docsgpt.deploy.docker import DeployError, Docker, lan_ip, wait_healthy
 
@@ -350,6 +352,119 @@ def env(args, context: Optional[Context] = None) -> int:
     return 0
 
 
+def _stack_image(env: Mapping[str, str]) -> str:
+    """The image this install runs; the volume tars go through it, so nothing extra is pulled."""
+    tag = env.get("DOCSGPT_IMAGE_TAG") or "latest"
+    return f"arc53/docsgpt:{tag}{env.get('DOCSGPT_IMAGE_VARIANT', '')}"
+
+
+def backup(args, context: Optional[Context] = None) -> int:
+    """Write a dump of the database and a tar of each data volume into one archive."""
+    context = context or Context.default(args)
+    directory = stack.stack_dir(args.dir)
+    env = _installed(directory)
+    if env is None:
+        return 1
+
+    out_dir = Path(args.out).expanduser() if args.out else directory / "backups"
+    taken_at = datetime.now(timezone.utc)
+    target = out_dir / backup_format.archive_name(taken_at)
+    image = _stack_image(env)
+
+    with tempfile.TemporaryDirectory() as workspace:
+        work = Path(workspace)
+        dump = work / backup_format.DUMP
+        print("Dumping the database ...")
+        with dump.open("w", encoding="utf-8") as handle:
+            context.docker.compose(
+                directory, "exec", "-T", "postgres",
+                "pg_dump", "--clean", "--if-exists", "-U", "docsgpt", "-d", "docsgpt",
+                stdout=handle,
+            )
+        volume_tars = {}
+        for name in backup_format.DATA_VOLUMES:
+            print(f"Archiving the {name} volume ...")
+            tar_path = work / f"{name}.tar"
+            context.docker.export_volume(f"{PROJECT}_{name}", tar_path, image)
+            volume_tars[name] = tar_path
+        manifest = {
+            "format": backup_format.FORMAT,
+            "created_at": taken_at.isoformat(timespec="seconds"),
+            "version": context.version,
+            "image_tag": env.get("DOCSGPT_IMAGE_TAG", ""),
+            "volumes": list(backup_format.DATA_VOLUMES),
+            "settings_included": bool(args.with_settings),
+        }
+        settings = directory / ".env" if args.with_settings else None
+        backup_format.write_archive(target, dump=dump, volume_tars=volume_tars, manifest=manifest, settings=settings)
+
+    size = target.stat().st_size / 1_000_000
+    print(f"\nBackup written to {target} ({size:.1f} MB)")
+    if args.with_settings:
+        print("It contains .env, so it holds this install's secrets: keep it somewhere private.")
+    else:
+        print("Settings are not in it; `docsgpt backup --with-settings` includes .env, secrets and all.")
+    print(f"Restore it with: docsgpt restore {target}")
+    return 0
+
+
+def restore(args, context: Optional[Context] = None) -> int:
+    """Put a backup's database and data volumes back over this install."""
+    context = context or Context.default(args)
+    archive = Path(args.archive).expanduser()
+    manifest = backup_format.read_manifest(archive)
+    backup_format.check_version(manifest, context.version, args.force)
+
+    directory = stack.stack_dir(args.dir)
+    env = _installed(directory)
+    if env is None:
+        return 1
+
+    taken_at = manifest.get("created_at", "an unknown time")
+    if not args.yes:
+        if not context.interactive:
+            raise DeployError("restore needs --yes when there is no terminal to confirm on")
+        question = f"Replace the data in {directory} with the backup from {taken_at}? This cannot be undone."
+        if not context.prompter.confirm(question, default=False):
+            print("Nothing restored.")
+            return 1
+
+    image = _stack_image(env)
+    print("Stopping the stack ...")
+    context.docker.compose(directory, *_EVERY_PROFILE, "down")
+
+    with tempfile.TemporaryDirectory() as workspace:
+        work = Path(workspace)
+        backup_format.extract(archive, work)
+        for name in manifest.get("volumes", []):
+            tar_path = work / backup_format.volume_member(name)
+            if not tar_path.is_file():
+                raise DeployError(f"{archive} is missing the {name} volume it says it contains")
+            print(f"Restoring the {name} volume ...")
+            context.docker.import_volume(f"{PROJECT}_{name}", tar_path, image)
+
+        dump = work / backup_format.DUMP
+        if not dump.is_file():
+            raise DeployError(f"{archive} is missing its database dump")
+        print("Starting the database ...")
+        context.docker.compose(directory, "up", "-d", "--wait", "postgres")
+        print("Restoring the database ...")
+        with dump.open("r", encoding="utf-8") as handle:
+            context.docker.compose(
+                directory, "exec", "-T", "postgres",
+                "psql", "--quiet", "-U", "docsgpt", "-d", "docsgpt",
+                stdin=handle,
+            )
+
+    print("Starting DocsGPT ...")
+    context.docker.compose(directory, "up", "-d", "--remove-orphans")
+    if not context.wait(stack.health_url(env), args.timeout):
+        print("DocsGPT did not answer after the restore. See `docsgpt logs backend`.", file=sys.stderr)
+        return 1
+    print(f"\nRestored the backup from {taken_at}. DocsGPT is running at {stack.url(env, context.lan_ip())}")
+    return 0
+
+
 def _uninstall_hint(installer: str) -> str:
     return {"uv": "uv tool uninstall docsgpt", "pipx": "pipx uninstall docsgpt"}.get(installer, "pip uninstall docsgpt")
 

docsgpt/deploy/docker.py

@@ -22,10 +22,16 @@ class DeployError(Exception):
     """A problem the user can act on; the command prints it without a traceback."""
 
 
-def run(args: Sequence[str], *, cwd: Optional[Path] = None, capture: bool = False, check: bool = True):
-    """Run a command, streaming its output unless ``capture``; with ``check`` a failure raises DeployError."""
+def run(args: Sequence[str], *, cwd: Optional[Path] = None, capture: bool = False, check: bool = True,
+        stdout=None, stdin=None):
+    """Run a command, streaming its output unless ``capture`` or a file is given for ``stdout``.
+
+    ``stdout`` and ``stdin`` take open files, so a dump goes straight to disk and
+    back in again without passing through this process.
+    """
+    streams = {"capture_output": True} if capture else {"stdout": stdout, "stdin": stdin}
     try:
-        result = subprocess.run(list(args), cwd=cwd, text=True, capture_output=capture, check=False)
+        result = subprocess.run(list(args), cwd=cwd, text=True, check=False, **streams)
     except FileNotFoundError as exc:
         raise DeployError(f"{args[0]} is not installed or not on PATH") from exc
     if check and result.returncode != 0:
@@ -94,9 +100,29 @@ class Docker:
             raise DeployError("Docker is not running. Start it with `sudo systemctl start docker` and run this again.")
         raise DeployError("Docker is not running. Start Docker Desktop and run this again.")
 
-    def compose(self, directory: Path, *args: str, capture: bool = False, check: bool = True):
+    def compose(self, directory: Path, *args: str, capture: bool = False, check: bool = True,
+                stdout=None, stdin=None):
         """``docker compose <args>`` in ``directory``, which holds the Compose file and its ``.env``."""
-        return self._run(["docker", "compose", *args], cwd=directory, capture=capture, check=check)
+        return self._run(["docker", "compose", *args], cwd=directory, capture=capture, check=check,
+                         stdout=stdout, stdin=stdin)
+
+    def export_volume(self, volume: str, dest: Path, image: str) -> None:
+        """Write ``volume`` to ``dest`` as a tar, through an image the install already has."""
+        dest.parent.mkdir(parents=True, exist_ok=True)
+        with dest.open("wb") as handle:
+            self._run(
+                ["docker", "run", "--rm", "-v", f"{volume}:/data:ro", image, "tar", "cf", "-", "-C", "/data", "."],
+                stdout=handle,
+            )
+
+    def import_volume(self, volume: str, source: Path, image: str) -> None:
+        """Replace ``volume``'s contents with the tar at ``source``; the volume is created when missing."""
+        with source.open("rb") as handle:
+            self._run(
+                ["docker", "run", "--rm", "-i", "-v", f"{volume}:/data", image,
+                 "sh", "-c", "find /data -mindepth 1 -delete && tar xf - -C /data"],
+                stdin=handle,
+            )
 
     def volume_exists(self, name: str) -> bool:
         return self._run(["docker", "volume", "inspect", name], capture=True, check=False).returncode == 0

tests/deploy/test_backup.py

@@ -0,0 +1,131 @@
+"""`docsgpt backup` and `docsgpt restore`, against a fake Docker."""
+
+import io
+import json
+import tarfile
+
+import pytest
+
+from docsgpt.deploy import backup as backup_module
+from docsgpt.deploy.docker import DeployError
+
+from .test_commands import FakeDocker, FakePrompter, _context, _run
+
+
+def _installed(tmp_path, *extra):
+    assert _run(["up", "--yes", "--dir", str(tmp_path), *extra], _context()) == 0
+
+
+def _archive_names(path):
+    with tarfile.open(path, "r:gz") as archive:
+        return sorted(member.name for member in archive.getmembers() if member.isfile())
+
+
+def _manifest(path):
+    with tarfile.open(path, "r:gz") as archive:
+        return json.loads(archive.extractfile(backup_module.MANIFEST).read())
+
+
+def _copy_with_version(archive, version, dest):
+    """The same archive, with another DocsGPT version written into its manifest."""
+    with tarfile.open(archive, "r:gz") as source, tarfile.open(dest, "w:gz") as out:
+        for member in source.getmembers():
+            body = source.extractfile(member).read() if member.isfile() else None
+            if member.name == backup_module.MANIFEST:
+                manifest = json.loads(body.decode())
+                manifest["version"] = version
+                body = json.dumps(manifest).encode()
+                member.size = len(body)
+            out.addfile(member, io.BytesIO(body) if body is not None else None)
+    return dest
+
+
+class TestBackup:
+    def test_writes_a_dump_a_tar_per_volume_and_a_manifest(self, tmp_path, capsys):
+        _installed(tmp_path)
+        docker = FakeDocker(volumes={"docsgpt_postgres_data"})
+        out = tmp_path / "backups"
+        assert _run(["backup", "--dir", str(tmp_path), "--out", str(out)], _context(docker)) == 0
+
+        archives = list(out.glob("docsgpt-*.tar.gz"))
+        assert len(archives) == 1, archives
+        names = _archive_names(archives[0])
+        assert backup_module.DUMP in names
+        for volume in ("indexes", "inputs", "vectors"):
+            assert f"volumes/{volume}.tar" in names
+        manifest = _manifest(archives[0])
+        assert manifest["version"] == "0.21.0"
+        assert manifest["image_tag"] == "0.21.0"
+        assert manifest["volumes"] == ["indexes", "inputs", "vectors"]
+        assert manifest["settings_included"] is False
+        assert "Backup written to" in capsys.readouterr().out
+        assert [op for op in docker.volume_ops if op[0] == "export"], docker.volume_ops
+
+    def test_the_settings_file_is_left_out_unless_asked_for(self, tmp_path):
+        _installed(tmp_path)
+        out = tmp_path / "backups"
+        assert _run(["backup", "--dir", str(tmp_path), "--out", str(out)], _context()) == 0
+        assert backup_module.SETTINGS not in _archive_names(next(out.glob("*.tar.gz")))
+
+        with_settings = tmp_path / "with-settings"
+        argv = ["backup", "--dir", str(tmp_path), "--out", str(with_settings), "--with-settings"]
+        assert _run(argv, _context()) == 0
+        assert backup_module.SETTINGS in _archive_names(next(with_settings.glob("*.tar.gz")))
+
+    def test_the_database_is_dumped_from_the_running_container(self, tmp_path):
+        _installed(tmp_path)
+        docker = FakeDocker()
+        assert _run(["backup", "--dir", str(tmp_path), "--out", str(tmp_path / "b")], _context(docker)) == 0
+        dumps = [args for _, args in docker.calls if "pg_dump" in " ".join(args)]
+        assert dumps, docker.calls
+        assert dumps[0][:3] == ["exec", "-T", "postgres"]
+
+    def test_without_an_install(self, tmp_path, capsys):
+        assert _run(["backup", "--dir", str(tmp_path)], _context()) == 1
+        assert "docsgpt up" in capsys.readouterr().err
+
+
+class TestRestore:
+    def _backup(self, tmp_path, *extra):
+        _installed(tmp_path)
+        out = tmp_path / "backups"
+        assert _run(["backup", "--dir", str(tmp_path), "--out", str(out), *extra], _context()) == 0
+        return next(out.glob("*.tar.gz"))
+
+    def test_restores_the_volumes_and_the_database(self, tmp_path):
+        archive = self._backup(tmp_path)
+        docker = FakeDocker(volumes={"docsgpt_postgres_data"})
+        argv = ["restore", str(archive), "--dir", str(tmp_path), "--yes"]
+        assert _run(argv, _context(docker)) == 0
+        joined = [" ".join(args) for _, args in docker.calls]
+        assert any(call.startswith("--profile https down") for call in joined), joined
+        assert any("psql" in call for call in joined), joined
+        assert any(call.startswith("up -d") for call in joined), joined
+
+    def test_asks_before_replacing_data(self, tmp_path):
+        archive = self._backup(tmp_path)
+        docker = FakeDocker(volumes={"docsgpt_postgres_data"})
+        prompter = FakePrompter([False])
+        assert _run(["restore", str(archive), "--dir", str(tmp_path)], _context(docker, prompter, interactive=True)) == 1
+        assert docker.calls == []
+
+    def test_refuses_an_archive_that_is_not_a_docsgpt_backup(self, tmp_path):
+        stray = tmp_path / "stray.tar.gz"
+        with tarfile.open(stray, "w:gz") as archive:
+            note = tmp_path / "note.txt"
+            note.write_text("not a backup")
+            archive.add(note, arcname="note.txt")
+        with pytest.raises(DeployError, match="not a DocsGPT backup"):
+            _run(["restore", str(stray), "--dir", str(tmp_path), "--yes"], _context())
+
+    def test_a_missing_archive(self, tmp_path):
+        with pytest.raises(DeployError, match="does not exist"):
+            _run(["restore", str(tmp_path / "nope.tar.gz"), "--dir", str(tmp_path), "--yes"], _context())
+
+    def test_a_newer_backup_is_refused_without_force(self, tmp_path):
+        """Restoring a 0.22 backup into 0.21 would hand an older schema newer data."""
+        newer = _copy_with_version(self._backup(tmp_path), "0.22.0", tmp_path / "newer.tar.gz")
+        with pytest.raises(DeployError, match="newer"):
+            _run(["restore", str(newer), "--dir", str(tmp_path), "--yes"], _context(FakeDocker()))
+        argv = ["restore", str(newer), "--dir", str(tmp_path), "--yes", "--force"]
+        assert _run(argv, _context(FakeDocker())) == 0

tests/deploy/test_commands.py

@@ -1,8 +1,10 @@
 """`docsgpt up` and the commands that manage the stack, against a fake Docker."""
 
+import io
 import json
 import subprocess
 import sys
+import tarfile
 from pathlib import Path
 
 import pytest
@@ -19,17 +21,38 @@ class FakeDocker:
         self.volumes = set(volumes)
         self.dirs = {Path(d) for d in project_dirs}
         self.calls = []
+        self.volume_ops = []
         self.preflights = 0
 
     def preflight(self, interactive=False):
         self.preflights += 1
 
-    def compose(self, directory, *args, capture=False, check=True):
+    def compose(self, directory, *args, capture=False, check=True, stdout=None, stdin=None):
         self.calls.append((Path(directory), list(args)))
         if "down" in args and "-v" in args:
             self.volumes.clear()
+        if stdout is not None:
+            stdout.write("-- fake pg_dump\n")
+        if stdin is not None:
+            self.restored_sql = stdin.read()
         return subprocess.CompletedProcess(["docker", "compose", *args], 0, stdout="", stderr="")
 
+    def export_volume(self, volume, dest, image):
+        """Write a small tar, as the real one does with `docker run ... tar cf -`."""
+        self.volume_ops.append(("export", volume, image))
+        dest = Path(dest)
+        dest.parent.mkdir(parents=True, exist_ok=True)
+        with tarfile.open(dest, "w") as tar:
+            body = f"{volume} contents".encode()
+            info = tarfile.TarInfo(f"{volume}.marker")
+            info.size = len(body)
+            tar.addfile(info, io.BytesIO(body))
+
+    def import_volume(self, volume, source, image):
+        self.volume_ops.append(("import", volume, image))
+        self.volumes.add(volume)
+        assert Path(source).is_file(), source
+
     def volume_exists(self, name):
         return name in self.volumes
 

tests/deploy/test_docker.py

@@ -15,9 +15,11 @@ class FakeRunner:
     def __init__(self, answers=None):
         self.answers = list((answers or {}).items())
         self.calls = []
+        self.streams = []
 
-    def __call__(self, args, *, cwd=None, capture=False, check=True):
+    def __call__(self, args, *, cwd=None, capture=False, check=True, stdout=None, stdin=None):
         self.calls.append((list(args), cwd))
+        self.streams.append((stdout, stdin))
         for prefix, answer in self.answers:
             if list(args[: len(prefix)]) == list(prefix):
                 if callable(answer):
@@ -105,6 +107,32 @@ class TestQueries:
         assert "label=com.docker.compose.project=docsgpt" in args
 
 
+class TestVolumes:
+    def test_export_writes_the_volume_through_the_stack_image(self, tmp_path):
+        runner = FakeRunner()
+        dest = tmp_path / "volumes" / "inputs.tar"
+        _docker(runner).export_volume("docsgpt_inputs", dest, "arc53/docsgpt:0.21.0")
+        args, _ = runner.calls[0]
+        assert args[:4] == ["docker", "run", "--rm", "-v"]
+        assert args[4] == "docsgpt_inputs:/data:ro"
+        assert args[5] == "arc53/docsgpt:0.21.0"
+        assert args[6:] == ["tar", "cf", "-", "-C", "/data", "."]
+        assert dest.is_file(), "the tar is written to the destination"
+        assert runner.streams[0][0] is not None, "stdout goes to the file, not through this process"
+
+    def test_import_replaces_the_volume_contents(self, tmp_path):
+        source = tmp_path / "inputs.tar"
+        source.write_bytes(b"tar")
+        runner = FakeRunner()
+        _docker(runner).import_volume("docsgpt_inputs", source, "arc53/docsgpt:0.21.0")
+        args, _ = runner.calls[0]
+        assert "docsgpt_inputs:/data" in args
+        assert args[-2] == "-c"
+        assert "tar xf - -C /data" in args[-1]
+        assert "find /data -mindepth 1 -delete" in args[-1], "old contents go first"
+        assert runner.streams[0][1] is not None, "the tar is fed in on stdin"
+
+
 class TestWaitHealthy:
     def test_succeeds_once_the_api_answers(self):
         attempts = {"n": 0}